Always Start with People, Not Technology

A people‑centred industrial strategy for AI, good work, and the everyday economy

People first should be a golden rule when thinking about economic development and its relationship to jobs and productivity. And yet, since the latest generative AI models launched at the start of the year, the conversation has been about technology and little else.

In “something big is happening” breathless narratives about the mass obsolescence of work — based on extreme scenarios, at least on the timescales asserted — we are missing the bigger picture of why we need to be creating better jobs in the everyday economy. The vast majority of people are not employed in places where frontier tech is likely to spread unimpeded, at least in the short term. They work in services — and pretty everyday services at that.

The top five occupations in the UK are retail sales assistants, care workers, managers, drivers, and teachers. Except for retail workers, none of these jobs are likely to be automated at any significant level any time soon. In fact, if we get technological adoption right, most will be augmented by AI — including retail workers. In a previous piece on AI and work, I outlined three broad categories: opportunity jobs, where AI enhances the role (such as lab scientist); at-risk jobs, where it replaces them (such as insurance clerk); and resilient jobs (such as teacher where AI may or may not be useful). The largest category by far is the last one.

This is where things get more complicated — and where a brilliant, (fairly) recent book from Dani Rodrik, Shared Prosperity in a Fractured World, becomes relevant. His argument is that we have been doing industrial policy all wrong, or at least that our approach is woefully incomplete. Policy has centred on manufacturing rather than on the service jobs that make up so much of the modern workforce.

In the US, after a decade of Trumpian protectionism followed by Bidenomics, employment in manufacturing actually declined from 8.5% to 8.1%. Bidenomics had many problems despite welcome proactivity in some sectors such as green industries; its political weakness was that it touched far too few American workers, so not enough foundational support was built — and, well, here we are. I explored this in Trump and the Everyday Economy:

“The Biden administration assumed that the politics would follow the policy. The reality, when it comes to economic development, is that the two have to march in tandem, feeding off one another. And given that so much of the workforce in distribution, construction, retail, hospitality, manufacturing, farming and property services sits outside of the high-growth sectors — albeit dependent upon them — the benefits of growth diffuse in ways often unrecognised by households. Their prism on politics is their direct experience: prices, pay and prospects.”

Rodrik’s argument — and he is entirely right — is that democracy depends on a strong middle class with good jobs, and that the decline of manufacturing has eroded exactly that. In the US, the share of adults living in middle-class households fell from 61% in 1971 to 50% in 2021, and life has become still more precarious given rising housing, education and health costs, and fragile savings.

The case he makes is that economic policy should be recalibrated to focus on the sectors that absorb the largest numbers of workers. Rodrik calls this new approach to industrial policy productivism — though that label doesn’t quite fit, given that the strategy moves away from a predominant focus on manufacturing. Peoplism might better capture what is proposed.

The core argument is that we should train our policy on raising the productivity of people in everyday economy sectors by a moderate amount. This won’t produce the spectacular productivity growth we saw through technology and organisational transformation during the postwar trente glorieuses — but it will be significant nonetheless, and it will enable millions of better jobs, thereby stabilising society and correcting the political economic weaknesses of decades of hyper-globalisation (though he is no protectionist), automation and austerity.

Which brings us back to AI. The alternative to a people-centred industrial strategy is a technology-centred one. As the latest wave of hyperbolic hype swirls, we are being implored — by former Prime Ministers and Chancellors of the Exchequer, no less — to adopt AI or be left behind. The history of technology adoption has shown the sceptics to be right in the short term and the boosters over the longer term. We just don’t know how long that longer term will be.

“Adopt or die” is a two-dimensional position. Even if AI adoption is compressed in time — and we are still only seeing transformational uptake in pockets — we have time to adjust and adapt deliberately. I do agree that we need to get on with it; but not through a desperate dash to inject US big-tech AI into every corner of the economy and public services, not least because of the risks posed by the current state of US governance and the ideological disposition of some behind the machine. There is still time for adaptive adoption.

It is easy to be distracted by technology. AI will clearly shift labour markets over time, and in the US there is already some evidence that it is doing so. We are beginning to see AI impacts on productivity growth there, as analysed by Erik Brynjolfsson and others. Five years into the generative AI revolution, its imprint on UK productivity and growth remains largely invisible — and very hard to distinguish from the noise of current labour market challenges, which appear to be more policy-related (rising costs) and cyclical, as post-pandemic hiring surges recede.

The problem with all this noise is that it is distracting us — and it is fuelling a chronic public loss of trust in AI, which in itself slows adoption. The more there is a hard sell, the more we pull back. The service-sector productivist critique of recent industrial policy seems more important than rapid AI adoption when it comes to supporting good work and a vibrant, growing middle-earning workforce. How we do that effectively is a question not just of speed but of approach. Any sensible good-work strategy would have three crucial elements: sovereign or public AI, a vibrant workertech innovation ecosystem, and a people-centred industrial policy.

Sovereign AI is critical to ensuring that AI systems are secure, do not create dependencies that conflict with our values, and are developed with public goals in mind — including supporting good work. A national programme is urgently needed. It is deeply problematic that the UK public sector is locking itself into hundreds of millions of pounds of investment in non-public or non-sovereign tech sitting at the critical nexus of security, defence and public services, when procurement could instead be used to build our own national capability.

Alongside this, we have almost no regulation — beyond health and safety law, employment law and data security — that shapes how technology is adopted in the workplace. New legislation is needed. I would propose an AI and Workers Act that would require employers to systematically and transparently consider how technology can augment work rather than replace it when adopting AI systems. Public AI could be at the heart of a research and innovation programme supporting this endeavour, exploring ways to enhance good work in employment-dense sectors.

The final piece of this new industrial policy jigsaw is a pivot towards a people-centred approach. Rodrik outlines much of this in Shared Prosperity. He describes a heavily adaptive programme in Wisconsin that has increased productivity through employer-led training with wraparound support for learners, including help with transport and childcare.

The key to productivity gains is enabling workers with less developed skills to take on new tasks and responsibilities — just as we have shifted some basic health checks from GP surgeries to pharmacies. This can save organisations and whole systems money while supporting improved wages. In Michigan, The Right Place provides comprehensive business support. Again, the trick is heavily targeted approaches, sustained over time, backed by a comprehensive array of support.

These programmes are reminiscent of the adaptive governance in China explored by Yuen Yuen Ang, and of regional growth strategies in South Korea covering everything from management skills to product innovation and supply chain development. The challenge for us now is to replicate these approaches across retail, care, construction, hospitality, real estate, and food and accommodation. This need not be ultra high-tech — it might be as simple as improving management skills, implementing a decent customer relationship management system, or providing basic worker training. There is a great deal of existing knowledge and technology just waiting to make a difference.

Get it right — alongside sovereign AI and workertech — and we will see good work spread geographically, because the everyday economy, unlike the frontier economy, is widely dispersed rather than concentrated. What could be better for restoring trust in our democracy and pushing back against the tide of populism and nativism?